Improving Filler/Suspect Similarity in Eyewitness Lineups Using Facial Recognition Systems

目击证人列队是破案的一个重要调查工具。在列队指认过程中，目击者看到一组6-8人，并被指示指出其中哪些人是犯罪者。阵容通常只包括一个人，警方认为是嫌疑人，和其他已知的无辜者称为填料。目击者对嫌疑犯的选择被视为有罪的证据，往往导致逮捕和起诉。因此，制定列队指认程序，最大限度地增加目击证人选择有罪嫌疑人的可能性，同时最大限度地减少他们错误地选择无辜嫌疑人的可能性，对公正司法至关重要。过去的研究表明，嫌疑人从填充物中脱颖而出的阵容，被称为有偏见的阵容，降低了识别准确性，主要是由于无辜嫌疑人的错误识别增加。嫌疑人与填充人非常相似，以至于很难与他们区分开来，这也会降低识别的准确性，使得这项任务对于其他可靠的证人来说太困难了。因此，目前的指导方针指示阵容建设者，以达到一个“甜蜜点”的相似性，嫌疑人足够相似的填料，但不太相似。然而，在实际实践中实施这些准则已被证明是具有挑战性的，因为没有普遍同意的测量填充物/可疑物相似性的方法，现有的研究也没有明确描述一种方法来实现可在该领域使用的相似性的“最佳点”。这个项目的主要目标是提高目击者阵容建设者的能力，包括适当水平的相似性，在他们的阵容填料。 该项目旨在开发有关填充物/嫌疑人相似性如何影响目击者识别准确性的基础知识，并在此过程中为新工具奠定基础，这些工具可以使阵容管理员更好地优化相似性以提高准确性。该项目将利用一种新兴技术--面部识别软件--被执法机构广泛采用来创建阵容，但迄今为止，目击者研究人员基本上没有对这种技术进行过研究。使用一个基于超过570，000张潜在填充物照片的基础数据库的新型实验平台，研究人员将进行两项研究，通过实验测量填充物/嫌疑人相似性-以其他研究人员和从业者广泛访问的方式通过算法测量-与目击者准确性的关系，以及与目击者准确性的关系如何与传统相似性措施进行比较。为了验证他们的发现，研究人员将进行一项前瞻性研究，将面部比较算法应用于一组新的阵容，以预测最有可能产生准确识别的人，然后评估这些预测的准确性。如果验证研究证明是成功的，该项目的研究结果可用于开发新的现场工具，使阵容管理员更容易实施有关填料/嫌疑人相似性的指导方针。更好地遵守这些指导方针可以提高阵容识别的质量和准确性，这反过来又会减少错误定罪的发生率。该奖项反映了NSF的法定使命，并被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

Eyewitness lineups are an essential investigative tool for solving crimes. During a lineup, eyewitnesses view a group of 6-8 individuals and are instructed to indicate which, if any, of the individuals is the perpetrator of the crime. Lineups typically include only one individual whom the police view as a suspect, and other known innocents called fillers. Selection of the suspect by the eyewitness is taken as evidence of guilt and often leads to arrest and prosecution. Thus, developing procedures for conducting lineups that maximize the likelihood that eyewitnesses will select guilty suspects while minimizing the likelihood that they will erroneously select innocent suspects is crucially important for the fair administration of justice. Past research suggests that lineups where the suspect stands out from the fillers, referred to as biased lineups, reduce identification accuracy, resulting primarily from an increase in false identifications of innocent suspects. Lineups in which the suspect resembles the fillers so closely as to be hard to distinguish from them also reduce identification accuracy, making the task too difficult for otherwise reliable witnesses. Current guidelines thus instruct lineup constructors to attain a "sweet spot" of similarity where suspects are similar enough to fillers, yet not too similar. Implementing these guidelines in actual practice, however, has proven challenging because there is no universally agreed upon way of measuring filler/suspect similarity, nor has existing research clearly delineated a method to achieve the "sweet spot" of similarity that can be used in the field. The main goal of this project is to enhance eyewitness lineup constructors' ability to include fillers with appropriate levels of similarity in their lineups. This project seeks to develop foundational knowledge about how filler/suspect similarity affects eyewitness identification accuracy, and, in doing so, lays the groundwork for new tools that could enable lineup administrators to better optimize similarity to improve accuracy. The project will exploit an emergent technology--facial recognition software--being widely embraced by law enforcement agencies to create lineups, but which has heretofore been largely unexamined by eyewitness researchers. Using a novel experimental platform built upon an underlying database of over 570,000 potential filler photographs, the investigators will conduct two studies that experimentally measure how filler/suspect similarity--measured algorithmically in a manner widely accessible to other researchers and practitioners--relates to eyewitness accuracy, and how its relationship with eyewitness accuracy compares to that of traditional similarity measures. To validate their findings, the investigators will then conduct a prospective study where they apply the facial comparison algorithm to a new set of lineups to predict those most likely to generate accurate identifications, and then assess the accuracy of these predictions. Should the validation study prove successful, the project's findings could be used to develop new field tools enabling lineup administrators to more easily implement guidelines regarding filler/suspect similarity. Better adherence to these guidelines can improve the quality and accuracy of lineup identifications which will, in turn, reduce the incidence of wrongful convictions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.